Cancer Classification Challenges in High-Dimensional Microarray Data: An In-Depth Exploration of Machine Learning Models
Malaysian Journal of Science Health & Technology · 14.03.2026 · 26-34
DOI: 10.33102/mjosht.531 ↗Kurz gefasst
Fachartikel in der Zeitschrift Malaysian Journal of Science Health & Technology, online erschienen am 14.03.2026. Verfasst von Wafa' Qasim Al-Jamal und 3 weiteren Autor:innen. Die Arbeit wurde bisher 14-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Microarray gene expression profiling has transformed biomedical research by enabling large-scale, parallel analysis of thousands of genes. Despite its promise, cancer classification using Machine Learning (ML) on microarray data continues to face critical challenges, particularly due to high dimensionality, limited sample sizes, and severe class imbalance. These factors contribute to overfitting, poor generalization, and inflated performance metrics, hindering the clinical translation of models. This Structured Literature Review (SLR) examines ML-based cancer classification studies published between 2015 and 2025. This period was marked by the emergence of deep learning, synthetic data generation, and biologically informed modeling. Using a transparent selection protocol, we synthesize findings from over 20 peer-reviewed studies. The review focuses on three methodological pillars: biologically grounded feature selection, constrained data augmentation, and robust performance evaluation. We identify a growing trend toward hybrid feature selection methods that balance statistical relevance and biological interpretability. However, comparative benchmarking across datasets remains limited. Data augmentation techniques, such as Synthetic Minority Oversampling Technique (SMOTE) and Generative Adversarial Networks (GAN)s, are increasingly being adopted. However, they often lack biological validation. This raises concerns about the plausibility of synthetic gene profiles. To address this, we recommend integrating pathway-level constraints and gene ontology checks during the augmentation process. Furthermore, we observe that many studies disproportionately emphasize accuracy. This can misrepresent the model's efficacy in imbalanced settings. Metrics such as Matthews Correlation Coefficient (MCC), F1-score, and precision-recall curves offer more reliable insights. These metrics should be standardized across evaluations. External validation using independent datasets is also essential to assess generalizability. In addition, it helps mitigate dataset-specific bias. Based on the findings, we present a conceptual hybrid framework that integrates biologically informed feature selection, biologically constrained data augmentation, and balanced evaluation protocols. This framework is intended to enhance reproducibility, biological fidelity, and translational reliability in machine learning-based cancer diagnostics, thereby contributing to the advancement of precision oncology. VIDEO EXPLAINER AUDIO PODCAST
Zitieren
APA
Al-Jamal, W. Q., Pitchay, S. A., Ridzuan, F., & Azam, M. H. N. (2026). Cancer Classification Challenges in High-Dimensional Microarray Data: An In-Depth Exploration of Machine Learning Models. Malaysian Journal of Science Health & Technology, 26-34. https://doi.org/10.33102/mjosht.531
Vancouver
Al-Jamal WQ, Pitchay SA, Ridzuan F, Azam MHN. Cancer Classification Challenges in High-Dimensional Microarray Data: An In-Depth Exploration of Machine Learning Models. Malaysian Journal of Science Health & Technology. 2026:26-34. doi:10.33102/mjosht.531
BibTeX
@article{AlJamal2026,
author = {Al-Jamal, Wafa' Qasim and Pitchay, Sakinah Ali and Ridzuan, Farida and Azam, Muhammad Harith Noor},
title = {Cancer Classification Challenges in High-Dimensional Microarray Data: An In-Depth Exploration of Machine Learning Models},
journal = {Malaysian Journal of Science Health & Technology},
year = {2026},
pages = {26--34},
doi = {10.33102/mjosht.531}
}
EUVIMED